CN105003249B - A horizontal well flow pattern identification method based on total flow and conductance probe array signals - Google Patents

Abstract

A horizontal well flow pattern identification method based on total flow and conductivity probe array signals belongs to the field of multiphase flow detection. First, measure the voltage response signal of each probe of the total flow and conductance probe array respectively; secondly, extract feature quantities from the voltage response signal of each probe by two techniques of statistical analysis and wavelet analysis; thirdly, analyze the extracted features The Z-score is normalized, and the principal components are extracted by principal component analysis (PCA) technology to become PCA feature quantities; then, feature-level information fusion based on support vector classification (SVC) is performed, that is, the SVC method is used to establish a The PCA characteristic quantities of the flow and probe array voltage response signals are used to classify the oil-water two-phase flow pattern. Finally, the parameters of the SVC model are optimized by the particle swarm optimization algorithm. The invention solves the problem that the central sampling device cannot identify the flow pattern of the horizontal well, greatly reduces the dimension of the input variable, and the addition of the total flow greatly improves the identification rate of the flow pattern of the horizontal well.

Description

A horizontal well flow pattern identification method based on total flow and conductance probe array signals

【Technical field】

The invention belongs to the field of multiphase flow detection, in particular to a horizontal well flow pattern identification method based on total flow and conductance probe array signals.

【Background technique】

Horizontal well technology is a new oilfield development technology developed in the 1920s. Due to its advantages of small production pressure difference and large oil drainage area, compared with vertical wells, it can greatly increase the oil production of a single well and the recovery of oil reservoirs. Yield, so it has received widespread attention in the field of oil exploration. Compared with vertical wells, my country's horizontal well technology is still very backward, so it is urgent to carry out research on horizontal well dynamic monitoring technology.

Flow pattern is an important parameter for multiphase flow parameter detection, which characterizes the distribution of each phase medium in the fluid flow process. In the study of two-phase flow, the flow characteristics and heat and mass transfer characteristics of the two-phase fluid are affected by the flow pattern, so the measurement of flow parameters is also affected by the flow pattern. Therefore, if the flow pattern in the well can be identified, a more suitable logging scheme can be selected to obtain a better logging effect. The flow pattern division map of 125mm inner diameter horizontal well based on CCD high-speed imaging method, divides the flow pattern into three layers: smooth stratified flow, stratified flow with mixture at the interface, continuous oil layer, continuous dispersed oil droplet layer and continuous water layer Flow (Reference: Jiang Changhua. Flow pattern analysis and visualization of oil-water two-phase flow in horizontal wells [D]. Beijing: Beihang University, 2013). At the same time, in order to cover the full scale of the water content of the experimental conditions, the oil single phase and the water single phase were identified together.

At present, the flow pattern identification of multiphase flow has been widely studied. Flow pattern identification methods include visual inspection, high-speed imaging, probe method, ray attenuation method, electrical tomography, amplitude domain processing method, time-frequency domain analysis method, information fusion method, nonlinear analysis method, etc. The State Intellectual Property Office has published and authorized a number of invention patents on flow pattern recognition. A published invention patent "Gas-liquid two-phase flow pattern identification method based on ICA and SVM" (application number 201410624191) uses differential pressure transmitter combined with independent component analysis (ICA) and support vector machine (SVM) to identify gas-liquid two-phase flow pattern. phase flow. An authorized invention patent "A Method for Identifying Two-Phase Flow Patterns Based on Hilbert's Marginal Spectrum" (Application No. 201110044591) uses electrostatic sensors to detect the flow noise signal of gas-solid two-phase flow, and then uses Hilbert Marginal spectrum analysis and neural network methods to identify gas-solid two-phase flow patterns. However, the method of the above invention cannot be applied to flow pattern identification of oil-water two-phase flow in production logging.

Conductivity probe method not only responds quickly to the change of flow parameters of oil-water two-phase flow, but also has low cost, safety, reliability and strong practicability, so it has been widely used. However, in highly deviated wells and horizontal wells, the multiphase fluids are separated due to gravity, resulting in uneven distribution of the medium, so that the central sampling device, such as a single probe located in the center, can only obtain information about the local fluid, and cannot measure multiple fluids. Phase flow parameters. In order to solve this problem, since the 1990s, internationally renowned oilfield service companies such as Schlumberger, Sondex and Computalog have successively developed logging instruments based on multi-probe structures, and have been tested in oil wells with large flow rates and large diameters. and application. Flores used the conductance probe array to identify the vertical and inclined oil-water two-phase flow respectively (reference Flores J.G. Oil-Water Flow in Vertical and Inclined Wells [D]. Tulsa: The University of Tulsa, 1997). The State Intellectual Property Office has authorized three invention patents related to the conductance probe array sensor and its optimization method, "A Multi-ring Electrode Array Imaging Sensor" (Patent No. ZL201010110504.0), "A Circular Water Holdup Logging Sensor Array" The structure optimization method of "(Patent No. ZL201010543247.X) and "a structure optimization method of multi-ring electrode array sensor based on genetic algorithm" (Patent No. ZL201210544383.X). However, the conductance probe method is far from mature, and the processing and use of the probe response signal still needs to be further studied. Combining soft-sensing methods with traditional multiphase flow sensors can greatly enrich the use of multiphase flow measurement data, thereby improving measurement accuracy. Generally, the soft measurement method includes the following steps: data mining, feature extraction, data fusion and parameter estimation. Therefore, it is very necessary to study the flow pattern identification method of horizontal well based on conductance probe array.

The oil-water two-phase flow distribution in the horizontal well depends on the total flow and water cut, and the total flow can be obtained by the turbine flowmeter after the flow is collected. If the total flow is used as a parameter to help describe the oil-water two-phase flow distribution, the identification rate of the flow pattern will be improved. According to the information level processed, the multi-sensor fusion system can be divided into three levels: data-level information fusion, feature-level information fusion and decision-level fusion. In a horizontal well, the oil-water two-phase flow medium is unevenly distributed, and a single probe cannot identify the flow pattern. It is necessary to study the voltage response signals of the conductance probe at different positions to identify the flow pattern. Therefore, the present invention adopts feature-level information fusion based on support vector classification.

The invention provides a horizontal well flow pattern identification method based on total flow and conductance probe array signals, which belongs to the field of multiphase flow detection. First, measure the voltage response signal of each probe of the total flow and conductance probe array respectively; secondly, extract feature quantities from the voltage response signal of each probe by two techniques of statistical analysis and wavelet analysis; thirdly, analyze the extracted features The Z-score is normalized, and principal component analysis (PCA) technology is used to extract the principal components to become PCA feature quantities; The PCA characteristic quantities of the flow and probe array voltage response signals are used to classify the oil-water two-phase flow pattern. Finally, the parameters of the SVC model are optimized by the particle swarm optimization algorithm. The invention solves the problem that the central sampling device cannot identify the flow pattern of the horizontal well, greatly reduces the dimension of the input variable, and the addition of the total flow greatly improves the identification rate of the flow pattern of the horizontal well.

[Content of the invention]

The purpose of the present invention is to provide a horizontal well flow pattern identification method based on total flow and conductance probe array signals to meet the requirements of production logging for high robustness, high reliability and high flow pattern identification rate.

In order to achieve the above purpose, the present invention provides a horizontal well flow pattern identification method based on total flow and conductance probe array signals, using the following technical solutions:

A horizontal well flow pattern identification method based on total flow and conductance probe array signals, characterized in that it comprises the following steps:

Step 1, under the combination of different total flow rates and water content of the oil-water two-phase flow in the horizontal well, the current collecting umbrella (25) is opened by the motor (24), and the total flow rate of the oil-water two-phase flow is measured by the turbine flowmeter (26);

In step 2, under the combination of different total flow rates and water content of the oil-water two-phase flow in the horizontal well, the support arm (222) of the conductance probe array (22) is opened by the motor (24), and the conductance probe is measured by the conductance measurement circuit (23) The voltage response signal of each probe (221) of the array (22), the measurement method is as follows, the bipolar sine wave excitation signal (31) whose amplitude is U _i is applied to the sampling resistor (32) whose resistance value is R _f , the switch (34) turns on each probe (35) of the conductance probe array in turn, the sampling resistance R _f and the position of the tip of the needle core (353) of the gated conductance probe are located between the oil-water two-phase flow (36). The resistance to ground _Rx constitutes a voltage divider circuit, and the amplitude of the voltage response signal (33) of the conductance probe measured at the peak moment of the excitation signal is U _o , then there is

The probe voltage response signal is recorded in the form of time series, and the measured data is stored by the storage and telemetry communication circuit (27), and compiled into a Mann code, which is connected to the logging cable through the cable interface (28) and uploaded to the ground;

Step 3: In the statistical analysis, four characteristic quantities are extracted from each probe voltage response signal, namely the mean, standard deviation, skewness coefficient, and kurtosis coefficient; in the wavelet analysis, the response time of each probe is The sequence is subjected to two-layer wavelet packet decomposition, and 8 feature quantities are extracted, that is, the energy ratio and information entropy of the four sub-band wavelet coefficients obtained by the second-layer wavelet decomposition; the method of extracting the feature quantities through wavelet analysis is as follows: The four sub-band wavelet coefficients obtained by layer wavelet decomposition can obtain the reconstruction sequence S _2,j of the corresponding sub-band, j=0,1,2,3; the four sub-band wavelet coefficients obtained by the second layer of wavelet decomposition Energy is

In the formula, S _2,j (k) represents the k-th element of the reconstruction sequence S _2,j , and N ₁ represents the length of S _2,j ; the energy ratio of the four sub-band wavelet coefficients obtained by the second-level wavelet decomposition It is calculated by the following formula

The information entropy of the four sub-band wavelet coefficients obtained in the second layer of wavelet decomposition is defined as

In the formula,

In the formula, _SF(2,j) (k) represents the kth element of the Fourier transform sequence of S2 _,j , and _N2 represents the length of _SF(2,j) .

In step 4, Z-score normalization is performed on the characteristic quantities of the voltage response signals of each probe of the conductance probe array, and then the principal component analysis (PCA) technique is used to extract the main characteristic quantities of the normalized characteristic quantities of all probes. component, to reduce the data redundancy between the feature quantities, and the obtained principal component is called the PCA feature quantity of the voltage response signal of the conductance probe array; the Z-score normalization method is defined as

表示归一化后的特征量向量，j＝1,2,…,N，N表示探针的数目，i＝1,2,…,12；μ_j,i和σ_j,i分别表示X_j,i的均值和标准差；PCA技术是分析多个变量间相关性的一种多元统计方法，通过正交变换将多个可能相关的变量变换成少数几个线性不相关的综合指标，称之为主成分，在所有正交变换线性组合中选取方差贡献率最高的综合指标作为第一主成分，后续的每个主成分都将是剩余线性组合中方差贡献率最高的综合指标，且与前面的主成分正交；In the above formula, X _j,i represents the vector composed of the i-th characteristic quantity of the j-th probe under different combinations of total flow rate and water content of the oil-water two-phase flow,

Represents the normalized feature vector, j=1,2,...,N, N represents the number of probes, i=1,2,...,12; μ _j,i and σ _j,i represent X _j , respectively _{, the mean and standard deviation of i} ; PCA technology is a multivariate statistical method to analyze the correlation between multiple variables. It transforms multiple possibly related variables into a few linearly uncorrelated comprehensive indicators through orthogonal transformation, which is called The principal component, the comprehensive index with the highest variance contribution rate is selected as the first principal component among all the linear combinations of orthogonal transformations, and each subsequent principal component will be the comprehensive index with the highest variance contribution rate in the remaining linear combinations, and it is the same as the previous one. The principal components of are orthogonal;

Step 5: Perform feature-level information fusion based on support vector classification (SVC) on the conductance probe array voltage response signal, that is, use the SVC method to establish the PCA feature quantity from the total flow and the conductance probe array voltage response signal to the horizontal well oil and water two. The identification model of the phase flow pattern is called the SVC model, and a sample of the training set is recorded as

(x _i ,y _i ), _xi ∈R ⁿ⁺¹ ,y _i ∈[1,5] (7) where x _i represents the n+1-dimensional input vector of the SVC model, where the n-dimensional input vector is the conductance PCA feature quantity of the probe array, n≤12×N, N represents the number of probes, and the other 1-dimensional input vector is the total flow measured by the turbine flowmeter; y _i represents the 1-dimensional output vector of the SVC model, which is the inner diameter of 125mm Horizontal well oil-water two-phase flow pattern, take 1 to represent smooth stratified flow, take 2 to represent stratified flow with mixture at the interface, take 3 to represent three-layer flow of continuous dispersed oil droplet layer and continuous water layer, take 4 to represent oil Single phase, take 5 to represent water single phase, i=1,2,...,l,l represents the length of the training set, the data format of the test set is the same as the training set; use the training set samples to train the SVC model, using Gaussian diameter To the basis function, use the test set samples to test the recognition rate of the horizontal well flow pattern of the SVC model;

In step 6, particle swarm optimization (PSO) algorithm is used to optimize the penalty factor C and the Gaussian radial basis function kernel radius σ of the SVC model, so as to improve the recognition rate and generalization ability of the SVC. The optimization steps are as follows: (a) Set the penalty factor C, the search range of the kernel function parameter σ, set the number of particles, the length of the particle, the range of the particle, the maximum speed of the particle, the learning factor, and the iteration termination condition. The iteration termination condition includes the maximum number of iterations and the intersection of the SVC model. According to the flow pattern recognition rate requirement under the verification, the position and velocity of the particle population are randomly initialized; (b) the fitness R _cv (C, σ) of each particle is calculated, that is, the flow pattern recognition rate of the horizontal well under the cross-validation of the SVC model; (c) In each iteration, the particle updates its speed and position by tracking the individual fitness extremum and the global fitness extremum, where the individual fitness extremum refers to the optimal fitness searched by the particle itself so far value, the global fitness extremum refers to the fitness optimal value found so far by the entire particle swarm; (d) if any one of the iteration termination conditions is reached, the iteration can be terminated, otherwise, return to step (b).

The invention solves the problem that the central sampling device cannot identify the flow pattern of the horizontal well, greatly reduces the dimension of the input variable, and the addition of the total flow greatly improves the identification rate of the flow pattern of the horizontal well.

【Instruction drawings】

Fig. 1 is a flow chart of a method for identifying flow patterns in horizontal wells based on total flow and conductance probe array signals;

Fig. 2 is a schematic diagram of an invasive retractable double-ring conductance probe array and a turbine flowmeter combined logging tool, in the figure, the centralizer (21), the conductance probe array (22), the conductance probe (221), and the support arm (222) ), conductance measurement circuit (23), motor (24), current collecting umbrella (25), turbine flowmeter (26), storage and telemetry communication circuit (27), cable interface (28);

Fig. 3 is a schematic diagram of the conductance measurement circuit measuring the voltage response signal of each probe of the conductance probe array, the bipolar sine wave excitation signal (31) in the figure, the sampling resistor (32) whose resistance value is _Rf , the conductance probe voltage response Signal (33), switch (34), conductivity probe (35), metal casing (351), insulating layer (352), needle core (353), horizontal well oil-water two-phase flow (36).

【Specific implementation plan】

With reference to Figures 1, 2 and 3, specific embodiments of the present invention will be further described with reference to examples.

In order to verify the invented method for identifying the horizontal well flow pattern based on the total flow and conductance probe array signals as shown in FIG. The combined logging tool carried out the oil-water two-phase flow experiment in the large-scale horizontal well multiphase flow experimental facility of the Daqing Petroleum Well Logging and Testing Experiment Center. The inner diameter of the horizontal simulated well is 125mm and the length is 16m. The double-loop conductance probe array logging tool consists of a centralizer (21), a conductance probe array (22), a conductance measurement circuit (23), a motor (24), a current collecting umbrella (25), a turbine flowmeter (26), a storage It is composed of a telemetry communication circuit (27) and a cable interface (28). The centralizer (21) can ensure that the logging instrument is in the center position in the wellbore. The 24 conductance probes (221) of the double-loop conductance probe array are equiangularly distributed on two circles concentric with the central axis of the logging tool, in a radial shape, and the two conductance probes on the same support arm (222) are parallel to each other . Each conductivity probe is composed of a metal casing (351), an insulating layer (352), and a needle core (353). The diameter of the metal casing (351) is 3mm, the casing is grounded, and the exposed tip of the needle core (353) is 3mm long. The insulating layer (352) Separate the needle core (353) from the metal shell (351), as shown in FIG. 3 . Each conductance probe can detect oil bubbles or water bubbles with a diameter greater than 3 mm through the conductance measurement circuit (23) and is not affected by the continuous phase, as shown in FIG. 3 . A motor (24) can open and retract the probe array (22) and the current collecting umbrella (25). When the collecting umbrella (25) is opened, the oil-water two-phase flow can be collected so as to facilitate the measurement of the total flow by the turbine flowmeter (26). The storage and telemetry communication circuit (27) can store the measured data, compile it into Mann code, and connect the logging cable to the surface through the cable interface (28).

The experimental oil was diesel oil, with a density of 0.825 g/cm ³ , a viscosity of 3×10 ^-3 Pa·s, and a surface tension of 28.62×10 ^-3 N/m. The water used was tap water, the density was 1 g/cm ³ , the viscosity was 0.890×10 ^-3 Pa·s, and the surface tension was 71.25×10 ^-3 N/m. In the experiment, the total flow rate of oil-water two-phase flow was set at 10-200 m ³ /day (adjustment interval 10 m ³ /day), and the water content was 0-100% (adjustment interval 10%). For various combinations of total flow and water cut, the 24 probes of the double-loop conductance probe array logging tool will record the voltage response signals of the conductance probes to obtain a measurement sample. Since there are 220 combinations of total flow and moisture content, 220 response signal samples were obtained for each probe. The sampling rate of each probe response signal was 0.1 kHz, and the length of each sample was 1300. In modeling, 220 samples of probe-response voltage were randomly divided into training set and test set, which accounted for 80% and 20% of the total samples, respectively. Repeat the random division process 50 times to obtain 50 combinations of training and test sets. These combinations were used to evaluate the method proposed by the present invention in a statistical sense.

A horizontal well flow pattern identification method based on total flow and conductance probe array signals, characterized in that it comprises the following steps:

Step 1, under the combination of different total flow rates and water content of the oil-water two-phase flow in the horizontal well, the current collecting umbrella (25) is opened by the motor (24), and the total flow rate of the oil-water two-phase flow is measured by the turbine flowmeter (26);

In step 2, under the combination of different total flow rates and water content of the oil-water two-phase flow in the horizontal well, the support arm (222) of the conductance probe array (22) is opened by the motor (24), and the conductance probe is measured by the conductance measurement circuit (23) The voltage response signal of each probe (221) of the array (22), the measurement method is as follows, the bipolar sine wave excitation signal (31) whose amplitude is U _i is applied to the sampling resistor (32) whose resistance value is R _f , the switch (34) turns on each probe (35) of the conductance probe array in turn, the sampling resistance R _f and the position of the tip of the needle core (353) of the gated conductance probe are located between the oil-water two-phase flow (36). The resistance to ground _Rx constitutes a voltage divider circuit, and the amplitude of the voltage response signal (33) of the conductance probe measured at the peak moment of the excitation signal is U _o , then there is

The probe voltage response signal is recorded in the form of time series, and the measured data is stored by the storage and telemetry communication circuit (27), and compiled into a Mann code, which is connected to the logging cable through the cable interface (28) and uploaded to the ground;

Step 3: In the statistical analysis, four characteristic quantities are extracted from each probe voltage response signal, namely the mean, standard deviation, skewness coefficient, and kurtosis coefficient; in the wavelet analysis, the response time of each probe is The sequence is subjected to two-layer wavelet packet decomposition, and 8 feature quantities are extracted, that is, the energy ratio and information entropy of the four sub-band wavelet coefficients obtained by the second-layer wavelet decomposition; the method of extracting the feature quantities through wavelet analysis is as follows: The four sub-band wavelet coefficients obtained by layer wavelet decomposition can obtain the reconstruction sequence S _2,j of the corresponding sub-band, j=0,1,2,3; the four sub-band wavelet coefficients obtained by the second layer of wavelet decomposition Energy is

In the formula, S _2,j (k) represents the k-th element of the reconstruction sequence S _2,j , and N ₁ represents the length of S _2,j ; the energy ratio of the four sub-band wavelet coefficients obtained by the second-level wavelet decomposition It is calculated by the following formula

The information entropy of the four sub-band wavelet coefficients obtained in the second layer of wavelet decomposition is defined as

In the formula,

In the formula, _SF(2,j) (k) represents the kth element of the Fourier transform sequence of S2 _,j , and _N2 represents the length of _SF(2,j) .

In step 4, Z-score normalization is performed on the characteristic quantities of the voltage response signals of each probe of the conductance probe array, and then the principal component analysis (PCA) technique is used to extract the main characteristic quantities of the normalized characteristic quantities of all probes. component, to reduce the data redundancy between the feature quantities, and the obtained principal component is called the PCA feature quantity of the voltage response signal of the conductance probe array; the Z-score normalization method is defined as

表示归一化后的特征量向量，j＝1,2,…,N，N表示探针的数目，i＝1,2,…,12；μ_j,i和σ_j,i分别表示X_j,i的均值和标准差；PCA技术是分析多个变量间相关性的一种多元统计方法，通过正交变换将多个可能相关的变量变换成少数几个线性不相关的综合指标，称之为主成分，在所有正交变换线性组合中选取方差贡献率最高的综合指标作为第一主成分，后续的每个主成分都将是剩余线性组合中方差贡献率最高的综合指标，且与前面的主成分正交；In the above formula, X _j,i represents the vector composed of the i-th characteristic quantity of the j-th probe under different combinations of total flow rate and water content of the oil-water two-phase flow,

Represents the normalized feature vector, j=1,2,...,N, N represents the number of probes, i=1,2,...,12; μ _j,i and σ _j,i represent X _j , respectively _{, the mean and standard deviation of i} ; PCA technology is a multivariate statistical method to analyze the correlation between multiple variables. It transforms multiple possibly related variables into a few linearly uncorrelated comprehensive indicators through orthogonal transformation, which is called The principal component, the comprehensive index with the highest variance contribution rate is selected as the first principal component among all the linear combinations of orthogonal transformations, and each subsequent principal component will be the comprehensive index with the highest variance contribution rate in the remaining linear combinations, and it is the same as the previous one. The principal components of are orthogonal;

Step 5: Perform feature-level information fusion based on support vector classification (SVC) on the conductance probe array voltage response signal, that is, use the SVC method to establish the PCA feature quantity from the total flow and the conductance probe array voltage response signal to the horizontal well oil and water two. The identification model of the phase flow pattern is called the SVC model, and a sample of the training set is recorded as

(x _i ,y _i ), _xi ∈R ⁿ⁺¹ ,y _i ∈[1,5] (7)

In the formula, x _i represents the n+1-dimensional input vector of the SVC model, where the n-dimensional input vector is the PCA feature of the conductance probe array, n≤12×N, N represents the number of probes, and the other 1-dimensional input vector is The total flow measured by the turbine flowmeter; y _i represents the 1-dimensional output vector of the SVC model, which is the oil-water two-phase flow pattern in a horizontal well with an inner diameter of 125 mm. Take 1 to represent smooth stratified flow, and take 2 to represent stratification with mixture at the interface flow, take 3 to represent the three-layer flow of the continuous dispersed oil droplet layer and the continuous water layer, take 4 to represent the oil single phase, take 5 to represent the water single phase, i=1,2,...,l,lrepresent the length of the training set, The data format of the test set is the same as that of the training set; the SVC model is trained using the training set samples, the Gauss radial basis function is used, and the test set samples are used to test the horizontal well flow pattern recognition rate of the SVC model;

Step 6: Use the particle swarm optimization (PSO) algorithm to optimize the penalty factor C and the Gaussian radial basis function kernel radius σ of the SVC model to improve the recognition rate and generalization ability of the SVC. The optimization steps are as follows: (a) Set Determine the penalty factor C, the search range of the kernel function parameter σ, set the number of particles, the length of the particle, the range of the particle, the maximum speed of the particle, the learning factor, and the iteration termination condition. The iteration termination condition includes the maximum number of iterations and the cross-validation of the SVC model. The position and velocity of the particle population are initialized randomly; (b) the fitness R _cv (C, σ) of each particle is calculated, that is, the flow pattern recognition rate of the horizontal well under the cross-validation of the SVC model; ( c) In each iteration, the particle updates its speed and position by tracking the individual fitness extremum and the global fitness extremum, where the individual fitness extremum refers to the fitness optimal value searched by the particle itself so far , the global fitness extremum refers to the fitness optimal value found by the entire particle swarm so far; (d) if any one of the iteration termination conditions is reached, the iteration can be terminated, otherwise, return to step (b).

In a horizontal well, the multiphase fluid is separated due to the action of gravity, resulting in uneven distribution of the medium, so that the central sampling device can only obtain the information of the local fluid and cannot measure the parameters of the multiphase flow. The horizontal well flow pattern identification method of the needle array signal solves this problem. The 24 probes of the conductance probe array have up to 288 features of voltage response signals. If the SVC model takes the PCA feature quantity of the voltage response signal of the conductance probe array as input, when the cumulative variance contribution rate of the principal component is 65.34%, that is, when the number of PCA feature quantities is 10, the recognition rate of the test set reaches the highest, which is 82.27%±5.86 %. If the SVC model takes the total flow and PCA features as input, when the cumulative variance contribution rate of the principal components is 36.51%, that is, when the number of PCA features is 5, the recognition rate of the test set reaches the highest, and it is greatly improved to 94.86% ± 3.27% ( mean ± standard deviation).

Therefore, the present invention solves the problem that the central sampling device cannot identify the flow pattern of the horizontal well, greatly reduces the dimension of the input variable, and the addition of the total flow greatly improves the identification rate of the flow pattern of the horizontal well.

The above is only the basic scheme of the specific implementation method of the present invention, but the protection scope of the present invention is not limited to this. Any person familiar with the technical field within the technical scope disclosed by the present invention, conceivable changes or substitutions, all should be included within the protection scope of the present invention. Therefore, the protection scope of the present invention should be subject to the protection scope of the claims. All changes that come within the meaning and range of equivalency of the claims are intended to be embraced within the scope of the claims.

Claims (1)

1. a horizontal well flow pattern identification method based on total flow and conductivity probe array signal, is characterized in that, comprises the following steps: Step 1, under the combination of different total flow rates and water content of the oil-water two-phase flow in the horizontal well, the current collecting umbrella (25) is opened by the motor (24), and the total flow rate of the oil-water two-phase flow is measured by the turbine flowmeter (26); In step 2, under the combination of different total flow rates and water content of the oil-water two-phase flow in the horizontal well, the support arm (222) of the conductance probe array (22) is opened by the motor (24), and the conductance probe is measured by the conductance measurement circuit (23) The voltage response signal of each probe of the array (22), the measurement method is as follows, the bipolar sine wave excitation signal (31) whose amplitude is U _i is applied on the sampling resistor (32) whose resistance value is R _f , the switch ( 34) Gating each probe of the conductance probe array in turn, the sampling resistance _Rf and the ground resistance Rx of the oil-water two-phase flow (36) at the position of the tip of the needle core (353) of the gated _conductance probe are formed The voltage divider circuit measures the amplitude of the voltage response signal (33) of the conductance probe at the peak moment of the excitation signal as U _o , then there is

The probe voltage response signal is recorded in the form of time series, and the measured data is stored by the storage and telemetry communication circuit (27), and compiled into a Mann code, which is connected to the logging cable through the cable interface (28) and uploaded to the ground; Step 3: In the statistical analysis, four characteristic quantities are extracted from each probe voltage response signal, namely the mean, standard deviation, skewness coefficient, and kurtosis coefficient; in the wavelet analysis, the response time of each probe is The sequence is subjected to two-layer wavelet packet decomposition, and 8 feature quantities are extracted, that is, the energy ratio and information entropy of the four sub-band wavelet coefficients obtained by the second-layer wavelet decomposition; the method of extracting the feature quantities through wavelet analysis is as follows: The four sub-band wavelet coefficients obtained by layer wavelet decomposition can obtain the reconstruction sequence S _2,j of the corresponding sub-band, j=0,1,2,3; the four sub-band wavelet coefficients obtained by the second layer of wavelet decomposition Energy is

In the formula, S _2,j (k) represents the k-th element of the reconstruction sequence S _2,j , and N ₁ represents the length of S _2,j ; the energy ratio of the four sub-band wavelet coefficients obtained by the second-level wavelet decomposition It is calculated by the following formula

The information entropy of the four sub-band wavelet coefficients obtained in the second layer of wavelet decomposition is defined as

In the formula,

In the formula, _SF(2,j) (k) represents the kth element of the Fourier transform sequence of S2 _,j , and _N2 represents the length of _SF(2,j) ; In step 4, Z-score normalization is performed on the characteristic quantities of the voltage response signals of each probe of the conductance probe array, and then the principal component analysis (PCA) technique is used to extract the main characteristic quantities of the normalized characteristic quantities of all probes. component, to reduce the data redundancy between the feature quantities, and the obtained principal component is called the PCA feature quantity of the voltage response signal of the conductance probe array; the Z-score normalization method is defined as

In the above formula, X _j,i represents the vector composed of the i-th characteristic quantity of the j-th probe under different combinations of total flow rate and water content of the oil-water two-phase flow,

Represents the normalized feature vector, j=1,2,...,N, N represents the number of probes, i=1,2,...,12; μ _j,i and σ _j,i represent X _j , respectively _{, the mean and standard deviation of i} ; PCA technology is a multivariate statistical method to analyze the correlation between multiple variables. It transforms multiple possibly related variables into a few linearly uncorrelated comprehensive indicators through orthogonal transformation, which is called The principal component, the comprehensive index with the highest variance contribution rate is selected as the first principal component among all the linear combinations of orthogonal transformations, and each subsequent principal component will be the comprehensive index with the highest variance contribution rate in the remaining linear combinations, and it is the same as the previous one. The principal components of are orthogonal; Step 5: Perform feature-level information fusion based on support vector classification (SVC) on the conductance probe array voltage response signal, that is, use the SVC method to establish the PCA feature quantity from the total flow and the conductance probe array voltage response signal to the horizontal well oil and water two. The identification model of the phase flow pattern is called the SVC model, and a sample of the training set is recorded as (x _i ,y _i ), _xi ∈R ⁿ⁺¹ ,y _i ∈[1,5] (7) where x _i represents the n+1-dimensional input vector of the SVC model, where the n-dimensional input vector is the conductance PCA feature quantity of the probe array, n≤12×N, N represents the number of probes, and the other 1-dimensional input vector is the total flow measured by the turbine flowmeter; y _i represents the 1-dimensional output vector of the SVC model, which is the inner diameter of 125mm Horizontal well oil-water two-phase flow pattern, take 1 to represent smooth stratified flow, take 2 to represent stratified flow with mixture at the interface, take 3 to represent three-layer flow of continuous dispersed oil droplet layer and continuous water layer, take 4 to represent oil Single phase, take 5 to represent water single phase, i=1,2,...,l,l represents the length of the training set, the data format of the test set is the same as the training set; use the training set samples to train the SVC model, using Gaussian diameter To the basis function, use the test set samples to test the recognition rate of the horizontal well flow pattern of the SVC model; In step 6, particle swarm optimization (PSO) algorithm is used to optimize the penalty factor C and the Gaussian radial basis function kernel radius σ of the SVC model, so as to improve the recognition rate and generalization ability of the SVC. The optimization steps are as follows: (a) Set the penalty factor C, the search range of the Gaussian radial basis function kernel radius σ, set the number of particles, the length of the particle, the range of the particle, the maximum speed of the particle, the learning factor, and the iteration termination condition. The iteration termination condition includes the maximum number of iterations and the flow pattern recognition rate requirement under the cross-validation of the SVC model, randomly initialize the position and velocity of the particle population; (b) calculate the fitness R _cv (C, σ) of each particle, that is, the horizontal well flow under the cross-validation of the SVC model (c) In each iteration, the particle updates its speed and position by tracking the individual fitness extremum and the global fitness extremum, where the individual fitness extremum refers to the searched particle itself so far The fitness optimal value, the global fitness extremum refers to the fitness optimal value found so far by the entire particle swarm; (d) If any one of the iteration termination conditions is reached, the iteration can be terminated, otherwise, return to step (b).

Images